An ensemble machine learning approach for predicting groundwater storage for sustainable management of water resources

Predicting groundwater storage (GWS) is essential for sustainable water resource management, especially in regions with water scarcity. This study proposes an ensemble machine learning (EML) approach (i.e., Bagging, XGBoost, and CatBoost) leveraging the Landsat-derived parameters to forecast GWS due to the limited availability of field observations. This modeling framework captures complex relationships between socioeconomic and environmental variables and groundwater storage in Chittagong City. Multiple indices, including the normalized difference vegetation index (NDVI), normalized difference built-up index (NDBI), modified normalized difference water index (MNDWI), and urban heat island (UHI), were utilized in the models. A digital elevation model (DEM), nighttime light (NTL), and nearest distance to water bodies from streamline data were used to investigate their impact on GWS. The empirical Bayesian kriging (EBK) method was used to downscale the GWS and NTL data. The outputs of the models were evaluated using statistical indicators such as the coefficient of determination (R²), root mean square error (RMSE), and Willmott's indicator of agreement (WIA). The Bagging and CatBoost models had higher R² and lower RMSE values (R² > 0.965, RMSE <1.604 mm) during the summer, while the XGBoost and CatBoost models performed better (R² > 0.966, RMSE <1.686 mm) in the winter. The results demonstrated that the utilization of Landsat metrics had the potential to serve as the predictive factors for estimating GWS. The proposed modeling framework can be used to predict GWS in regions with limited data, which will help policymakers, urban planners, and environmental organizations develop sustainable groundwater management strategies.